preprocess

Code

# targets::tar_load(params2$ss)
# ss<-eval(rlang::sym(params2$ss))

Illumina Infinium Methylation Arrays

This technical guide covers how to analyze Illumina methylation Beadchips arrays commonly known as EPIC/450K methylation arrays. here some description about a methylation array and how it is prepeared…

Preprocessing

The first step once you have your data is to load it to R in an appropriate format. The ´minfi´ packages provides read.metharray functions for this purpose. Also the functions read.metharray.sheet, read.metharray.exp can be used. Those are wrappers around read.metharray to make it easier for the user. Check this guideline or type vignette("minfi) on an R console for more info. In this case a built in function cnv.methyl::read.metharray.exp.par() is used which is a wrapper for minfi formula with parallelization to speed up this process a little bit.

In order to use any of this formulas we first need to generate the sample_sheet.

Sample sheet

The sample sheet contains inforamtion about your samples and the experimental setup. It is mandatory to respect the column names in order to make the pipline work. ##It must contain the following columns: - Sample_Name - Basename

##Other recommended columns with technical details. Can be used to track and detect batch effects: - Project - Pool_ID - Sample_Plate - Sample_Well - Sample_Group - Sentrix_ID - Sentrix_Position

##Pheno columns: - Gender (will be predicted) - Type - Condition - Any other information

In the following table, you can take a look at the sample sheet used for this project.

Code

# 
# renv::install("circlize")
# renv::install("RColorBrewer")
# renv::install("DT")
# renv::install("ggplot2")
# renv::install("shiny")
# renv::install("shinyWidgets")
# renv::install("ggplot2")
# renv::install("data.table")
# renv::install("writexl")

library("circlize")

========================================
circlize version 0.4.15
CRAN page: https://cran.r-project.org/package=circlize
Github page: https://github.com/jokergoo/circlize
Documentation: https://jokergoo.github.io/circlize_book/book/

If you use it in published research, please cite:
Gu, Z. circlize implements and enhances circular visualization
  in R. Bioinformatics 2014.

This message can be suppressed by:
  suppressPackageStartupMessages(library(circlize))
========================================

Code

library("RColorBrewer")
library("DT")
library("ggplot2")
library("shiny")

Attaching package: 'shiny'

The following objects are masked from 'package:DT':

    dataTableOutput, renderDataTable

Code

library("shinyWidgets")
library("data.table")
library("writexl")

top_beta <- function(beta_values, n=1000){
  sdv <- apply(beta_values, 1, sd)
  top100 <- names(head(sort(sdv,decreasing=T), n))
  beta_top100 <- beta_values[top100,]
  return(beta_top100)
}

Code

dtable<-function(data){
  DT::datatable(
        { data},
        filter = 'top',
        # selection = list(mode = 'multiple', selected = c(1:10), target = 'column', selectable = c(-2, -3)),
        fillContainer = F,
        # style =  "bootstrap",

        extensions = 'Buttons',

        options = list(
          paging = TRUE,
          searching = TRUE,
          fixedColumns = TRUE,
          autoWidth = FALSE,
          scrollX=TRUE,
          digits=4,
          ordering = TRUE,
          dom = 'Bfrtip',
          buttons = list(
            list(
              extend = "collection",
              text = 'download entire dataset',
              action = DT::JS("function ( e, dt, node, config ) {
                                                    Shiny.setInputValue('test', true, {priority: 'event'});
                                                    }")
            ),
            'copy',
            'csv',
            'excel'
          ),

          class = "display",
          server=TRUE
        ),
      ) |> DT::formatRound(which(sapply(data,is.double)),4)
}

Code

myModal <- function() {

  div(id = "test",
      shiny::modalDialog(downloadButton("download1","Download data as csv"),
                         br(),
                         br(),
                         downloadButton("download2","Download data as excel"),
                         easyClose = TRUE, title = "Download Table")
  )
}

renderDT<- function(data){
  output$dtable <- DT::renderDataTable({
    dtable(data)
  })

    shiny::observeEvent(input$test, {
      print("hello")
      showModal(myModal())
    })
    output$download1 <- shiny::downloadHandler(
      filename = function() {
        paste("data-", Sys.Date(), ".csv", sep="")
      },
      content = function(file) {
        write.csv(data, file)
      }
    )

    output$download2 <- shiny::downloadHandler(
      filename = function() {
        paste("data-", Sys.Date(), ".xlsx", sep="")
      },
      content = function(file) {
        writexl::write_xlsx(data, file)
      })
}

Code

dtable(data.table::as.data.table(ss))

Heatmap

First of all let’s define some functions for the color scales our heatmap will use:

Color functions:

Code

library(ComplexHeatmap)
library(circlize)
# Simple continuous palette: 
col_fun = colorRamp2(c(-1,-0.1, 0.1, 1), c("blue","white","white","red"))

# Continuous, uses a predefined color palette or manual color vector
col_fn<- function(x,n=100,palette=viridis::cividis(n)){
  colorRamp2( seq(min(x,na.rm = T),max(x, na.rm = T),length.out=n), palette)
}

# Continuous, monochrome breakpoints based on values distribution (quartiles):
col_fq <- function(x,probs=c(0,0.25,0.5,0.75,1),color ){
  colorRamp2( quantile(x,probs=probs), 
              monochromeR::generate_palette("white",
                blend_colour = color,  n_colours = length(probs)
                )
  )
}

# Continuous, monochrome can manually set breakpoints:
col_fbp <- function(x,bp,color){
  colorRamp2( bp, 
              monochromeR::generate_palette("white",
                blend_colour = color,  n_colours = length(bp)
                )
              )
}

Heatmaps function

Here we define the function for the heat maps

Code

library(ComplexHeatmap)


meth_heatmap <- function(samplesheet, bvals, annotation, idcol="rn"   ){
  require(ComplexHeatmap)
  require(data.table)
  data.table::setDT(annotation)
  setkeyv(annotation,idcol)
  
  # Annotation object for top annotation:
  data.table::setDT(samplesheet)
  # purity = samplesheet[,.SD,.SDcols=startsWith(names(samplesheet),"purity")]
  ha_column = HeatmapAnnotation(
    annotation_name_side = "left", 
    Type = anno_block(gp=gpar(fill= rainbow(n=length(unique(samplesheet$Type)))),
                      labels = unique(samplesheet$Type), #unique(samplesheet$Type),
                      labels_gp = gpar(col = "white", fontsize = 10)),
    Condition = samplesheet$Condition,
    # purity = unlist(purity),

    col = #A list of named vectors were names = vector values and value = color.
      list(
      Condition = setNames(
        palette.colors(length(unique(samplesheet$Condition)),
                       palette = "Dark"
                       ),
        unique(samplesheet$Condition)
      ),
      # purity = col_fbp(x=samplesheet,bp=seq(min(purity),max(purity),length.out=5),color="green3"),
      NULL
      )
  )
  
  heat_list<-ComplexHeatmap::Heatmap(
      matrix = bvals, 
            #Color:
            col=col_fn(bvals),#col_fun, # Color defined in col_fun above
            na_col="grey",
            
            #Label:
            heatmap_legend_param = list(
            at = c(0, 0.5, 1),
            # labels = c("hypo", 0, "hyper"),
            title = paste0(expression(beta),"-vals"),
            legend_height = unit(4, "cm"),
            title_position = "leftcenter-rot"
            ),
            
            
            #Rows:
            show_row_names = F,
            #row_title = "Amino acids",
            row_names_side = "left",
            #left_annotation = ha_boxplot,
            # clustering_distance_rows = "manhattan",
            
            #Columns:
            show_column_names = TRUE,
            column_names_side = "top",
            column_title_side = "bottom",
            column_names_max_height = unit(4, "cm"),
            column_names_gp = gpar(fontsize = 9),
            column_names_rot = 90,
            cluster_columns=F,
            column_split = samplesheet$Type, #factor(samplesheet$Type,levels = c("Case","Control")),
            column_title = paste0(expression(beta)," values"),
  
            #Annotation bar:
            top_annotation = ha_column,
           
              
            
            #Aspect ratios:
            #column_dend_height=unit(4, "cm")
             heatmap_width = unit(2, "npc"),
            heatmap_height = unit(16, "cm"),
                )
  # Add methylation difference heatmap:  
  for (contrast in unique(annotation$Contrast))
  {

    DMPann <- annotation[ Contrast==contrast,]
    # make sure to have one and only one bval for each cg probe, if more than one do mean:
    setDT(DMPann)
    setkeyv(DMPann,idcol)
    # Remove all annotation but methylation difference in contrast and probeid:
    mat<-DMPann[rownames(bvals) ,.SD,on=idcol,.SDcols=c("diff_meanMeth",idcol)]
    # If more than one id (idcol) do the mean
    mat<-mat[,.(bval=mean(diff_meanMeth)),by=idcol]
    mat<-mat$bval
    heat_list = heat_list + Heatmap(
      matrix = mat, 
      name = contrast,
            #Color:
            col=col_fun,#col_fun, # Color defined in col_fun above
            na_col="white",
            
            #Legend:
            heatmap_legend_param = list(
            at = c(-1, -0.1, 0.1, 1),
            # labels = c("hypo", "", "", "hyper"),
            title = paste0(expression(beta),"diff"),
            legend_height = unit(4, "cm"),
            title_position = "leftcenter-rot"
            ),
      show_heatmap_legend = ifelse(contrast == unique(annotation$Contrast)[1],T,F ),
            
            #Columns:
            show_column_names = TRUE,
            column_names_side = "top",
            column_names_max_height = unit(2, "cm"),
            column_names_gp = gpar(fontsize = 9),
            column_names_rot = 90,
            heatmap_width = unit(2, "npc"),
            
                )
    
  }
  # Add annotation
  ann<-annotation[,.(V1=unique(Relation_to_Island)),by=idcol]
  
  mat<-ann[rownames(bvals),V1,on=idcol]
  heat_list<-heat_list + 
    Heatmap(
      matrix = mat, 
      name = "Relation to island",
      #Color:
      col=setNames(palette.colors(length(unique(annotation$Relation_to_Island)),palette = "Set1"),sort(unique(annotation$Relation_to_Island))),#col_fun, # Color defined in col_fun above
      #Legend:
      show_heatmap_legend = TRUE,
      show_column_names = TRUE,
      column_names_side = "top",
      column_names_max_height = unit(2, "cm"),
      column_names_gp = gpar(fontsize = 9),
      column_names_rot = 90,
      heatmap_width = unit(1, "npc"),
            
                )
} 

PDC11

Methylation data: The 2-channel methylation intensity file format .idat

Code

ss_PDC11 <- data.table::as.data.table(tar_read(ss_clean_PDC11_batch2))
betas_PDC11 <- tar_read(betas_PDC11_batch2)
colnames(betas_PDC11) <- ss_PDC11[colnames(betas_PDC11),Sample_Name,on="barcode"]

# Sort by type:
ss_PDC11 <- setorder(ss_PDC11,Type)
betas_PDC11<-betas_PDC11[,ss_PDC11$Sample_Name]

DMPs Annotation data:

A dataframe with differentially methylated probes and annotation.

Code

library(data.table)

DMPann_PDC11<-tar_read(dmps_mod1_PDC11_batch2)
DMPann_PDC11<-data.table::setDT(DMPann_PDC11)
idcol<-"EPICv1_Loci"
setkeyv(DMPann_PDC11,idcol)
PROBEIDS <- intersect (rownames(betas_PDC11), DMPann_PDC11[[idcol]])
betas_PDC11 <- betas_PDC11[PROBEIDS,]
DMPann_PDC11 <- DMPann_PDC11[PROBEIDS,]
#Transform to data.table format to visualize and perform data manipulations
DMPann_PDC11<-data.table::setDT(DMPann_PDC11)

Heatmap

Code

heat_list_PDC11<-meth_heatmap(samplesheet = ss_PDC11, bvals = betas_PDC11, annotation = DMPann_PDC11, idcol = idcol)
ComplexHeatmap::draw(heat_list_PDC11, row_title = paste0("top ",NROW(betas_PDC11)," across-sample most variable sites"), row_title_gp = gpar(col = "darkblue"),
    column_title = "PDC11 cell line Differentially Methylated Probes Heatmap", column_title_gp = gpar(fontsize = 16),  merge_legend = TRUE)

H358

Methylation data: The 2-channel methylation intensity file format .idat

Code

ss_H358 <- data.table::as.data.table(tar_read(ss_clean_H358))
betas_H358 <- tar_read(top_H358)
colnames(betas_H358) <- ss_H358[colnames(betas_H358),Sample_Name,on="barcode"]
# Sort by type:
setorder(ss_H358,Type)
betas_H358<-betas_H358[,ss_H358$Sample_Name]

DMPs Annotation data:

A dataframe with differentially methylated probes and annotation.

Code

library(data.table)

DMPann_H358<-tar_read(dmps_mod1_H358)
DMPann_H358<-data.table::setDT(DMPann_H358)
idcol<-"Name"
setkeyv(DMPann_H358,idcol)
PROBEIDS <- intersect (rownames(betas_H358), DMPann_H358[[idcol]])
betas_H358 <- betas_H358[PROBEIDS,]
DMPann_H358 <- DMPann_H358[PROBEIDS,]
#Transform to data.table format to visualize and perform data manipulations
DMPann_H358<-data.table::setDT(DMPann_H358)

Heatmap

Code

heat_list_H358<-meth_heatmap(samplesheet = ss_H358, bvals = betas_H358, annotation = DMPann_H358, idcol = idcol)
ComplexHeatmap::draw(heat_list_H358, row_title = paste0("top ",NROW(betas_H358)," across-sample most variable sites"), row_title_gp = gpar(col = "darkblue"),
    column_title = "H358 cell line Differentially Methylated Probes Heatmap", column_title_gp = gpar(fontsize = 16),  merge_legend = TRUE)

H2009

Methylation data: The 2-channel methylation intensity file format .idat

Code

ss_H2009 <- data.table::as.data.table(tar_read(ss_clean_H2009))
betas_H2009 <- tar_read(top_H2009)
colnames(betas_H2009) <- ss_H2009[colnames(betas_H2009),Sample_Name,on="barcode"]
# Sort by type:
setorder(ss_H2009,Type)
betas_H2009<-betas_H2009[,ss_H2009$Sample_Name]

DMPs Annotation data:

A dataframe with differentially methylated probes and annotation.

Code

library(data.table)

DMPann_H2009<-tar_read(dmps_mod1_H2009)
DMPann_H2009<-data.table::setDT(DMPann_H2009)
idcol<-"Name"
setkeyv(DMPann_H2009,idcol)
PROBEIDS <- intersect (rownames(betas_H2009), DMPann_H2009[[idcol]])
betas_H2009 <- betas_H2009[PROBEIDS,]
DMPann_H2009 <- DMPann_H2009[PROBEIDS,]
#Transform to data.table format to visualize and perform data manipulations
DMPann_H2009<-data.table::setDT(DMPann_H2009)

Heatmap

Code

heat_list_H2009<-meth_heatmap(samplesheet = ss_H2009, bvals = betas_H2009, annotation = DMPann_H2009, idcol = idcol)
ComplexHeatmap::draw(heat_list_H2009, row_title = paste0("top ",NROW(betas_H2009)," across-sample most variable sites"), row_title_gp = gpar(col = "darkblue"),
    column_title = "H2009 cell line Differentially Methylated Probes Heatmap", column_title_gp = gpar(fontsize = 16),  merge_legend = TRUE)

Save Heatmaps:

Plot size:

There are many things you can do in order to change the plot size and aesthetics. I won’t go in detail but I think that getting the right plot size can be a bit tricky. So here is some advice on how to get it right. Here is a function to get the width and hight of your plot.

Code

calc_ht_size = function(ht, unit = "inch") {
    pdf(NULL)
    ht = ComplexHeatmap::draw(ht)
    w = ComplexHeatmap:::width(ht)
    w = convertX(w, unit, valueOnly = TRUE)
    h = ComplexHeatmap:::height(ht)
    h = convertY(h, unit, valueOnly = TRUE)
    dev.off()

    c(w, h)
}

Code

size <- calc_ht_size(heat_list_PDC11)
pdf(paste0("PDC11_heatmap.pdf"), width = size[1], height = size[2])
heat_list_PDC11
dev.off()

png 
  2 

Code

size <- calc_ht_size(heat_list_H2009)
pdf(paste0("H2009_heatmap.pdf"), width = size[1], height = size[2])
heat_list_H2009
dev.off()

png 
  2 

Code

size <- calc_ht_size(heat_list_H358)
pdf(paste0("H358_heatmap.pdf"), width = size[1], height = size[2])
heat_list_H358
dev.off()

png 
  2 

(a) PDC11

(b) H358

Figure 1:

When the number of rows is high the image gets compressed and we can loose some information. Also, sometimes we have different heatmaps with a different number of rows for each of them, which results on different aspect ratios. If we want to control the height of each row and keep it stable between plots we must consider: - By convention resolution in R is 96, which means 96 pixels fit on an inch. So if you don’t want to loose any line your cell height should be 1 pixel tall at least. 96 lines in 1 inch means each line is 0.26mm which is already very small, so don’t try to make it smaller.

• Complexheatmat default top and bottom margins are 2mm so you must add those 4mm to your plot size.
• The relationship between main heatmap object and cell height is proportional: The height of the main plot is controlled by the height parameter the rest is annotation as explained in the docs. So let’s find the ratio for a given cell size:

Code

library(ComplexHeatmap)
plotsizes=c(100,200)
y = NULL
for(nr in plotsizes) {
  betas<-top_beta(betas_PDC11,nr)
  
    ht = ComplexHeatmap::draw(meth_heatmap(ss_PDC11,betas,annotation = DMPann_PDC11,idcol = "EPICv1_Loci"),height=unit(1/96, "inch")*nr)
    ht_height = sum(component_height(ht)) + unit(4, "mm")
    ht_height = convertHeight(ht_height, "inch", valueOnly = TRUE)
    y = c(y, ht_height)
}

Code

sizemod <- lm(y ~ plotsizes)
sizemod

Call:
lm(formula = y ~ plotsizes)

Coefficients:
(Intercept)    plotsizes  
    2.43104      0.01042  

Now we use the formula to control for the plot size, so we can make the plots from the different cell lines proportional in row height:

Code

 png(paste0("PDC11_heatmap_resized.png"),
     width =  14,
     height = sizemod$coefficients[2]*NROW(betas_PDC11) + sizemod$coefficients[1],
     units = "in",
     res = 96
     )
ComplexHeatmap::draw(heat_list_PDC11, row_title = paste0("top ",NROW(betas_PDC11) ," across-sample EPICv1 most variable sites"), row_title_gp = gpar(col = "darkblue") , height = unit(1*1/96, "inch")*NROW(betas_PDC11), column_title = "PDC11 cell line Differentially Methylated Probes Heatmap",  merge_legend = TRUE)

dev.off()

 png(paste0("H358_heatmap_resized.png"),
     width =  14,
     height = sizemod$coefficients[2]*NROW(betas_H358) + sizemod$coefficients[1]+2,
     units = "in",
     res = 96
     )
ComplexHeatmap::draw(heat_list_H358, row_title = paste0("top ",NROW(betas_H358) ," across-sample EPICv1 most variable sites"), row_title_gp = gpar(col = "darkblue") , height = unit(1*1/96, "inch")*NROW(betas_H358), column_title = "H358 cell line Differentially Methylated Probes Heatmap",  merge_legend = TRUE)

dev.off()


 png(paste0("H2009_heatmap_resized.png"),
     width =  14,
     height = sizemod$coefficients[2]*NROW(betas_H2009) + sizemod$coefficients[1],
     units = "in",
     res = 96
     )
ComplexHeatmap::draw(heat_list_H2009, row_title = paste0("top ",NROW(betas_H2009) ," across-sample EPICv1 most variable sites"), row_title_gp = gpar(col = "darkblue") , height = unit(1*1/96, "inch")*NROW(betas_H2009), column_title = "H2009 cell line Differentially Methylated Probes Heatmap",  merge_legend = TRUE)

dev.off()

(a) PDC11

H358

Figure 2:

21/11/2023 Custom plots REQUEST:

The objective now is to generate 4 plots.

Individual plots:

• 3 plots (1 plot x Cell Line) with the top 1000 variable sites within the cell line.

Combined plot:

• 1 plot with the top 3000 most variable sites across all the cell lines.

Function to select top n variable sites:

Code

top_beta <- function(beta_values, n=1000){
  sdv <- apply(beta_values, 1, sd)
  topn <- names(head(sort(sdv,decreasing=T), n))
  beta_topn <- beta_values[topn,]
  return(beta_topn)
}

Individual plots:

H2009

Load the data:

Code

ss_H2009 <- readRDS("data/ss_H2009.rds")[-1,]
betas_H2009 <- readRDS("data/betas_H2009.rds")
DMPann_H2009 <- readRDS("data/DMPann_H2009.rds")
idcol<-"Name"

Apply the function to select the top 1000 most variable sites:

Code

n = 1000                                               # change this value to select a different number of probes
betas_H2009_heat <- top_beta(betas_H2009,n=n)

Generate the plot using the functions defined in the Section 0.3 section:

Code

heat_list_H2009 <- meth_heatmap(samplesheet = ss_H2009, bvals = betas_H2009_heat, annotation = DMPann_H2009, idcol = idcol)

Save with the desired resolution and titles:

Code

library(ComplexHeatmap)
library(circlize)

ComplexHeatmap::draw(heat_list_H2009, row_title = paste0("top ",n ," across-sample most variable sites"), row_title_gp = gpar(col = "darkblue") , height = unit(1*1/96, "inch")*n, column_title = paste0( "H2009 cell line top ", n, " most variable Probes Heatmap"),  merge_legend = TRUE)

H358

First we need to load the data:

Code

ss_H358 <- readRDS("data/ss_H358.rds")[-1,]
betas_H358 <- readRDS("data/betas_H358.rds")
DMPann_H358 <- readRDS("data/DMPann_H358.rds")
idcol<-"Name"

Then apply the function to select the top 1000 most variable sites:

Code

n = 1000                                               # change this value to select a different number of probes
betas_H358_heat <- top_beta(betas_H358,n=n)

Now we can generate the plot this using the functions defined Section 0.3:

Code

heat_list_H358 <- meth_heatmap(samplesheet = ss_H358, bvals = betas_H358_heat, annotation = DMPann_H358, idcol = idcol)

And finally save with the desired resolution and titles:

Code

library(ComplexHeatmap)
library(circlize)

ComplexHeatmap::draw(heat_list_H358, row_title = paste0("top ",n ," across-sample most variable sites"), row_title_gp = gpar(col = "darkblue") , height = unit(1*1/96, "inch")*n, column_title = paste0( "H358 cell line top ", n, " most variable Probes Heatmap"),  merge_legend = TRUE)

PDC11_batch2

First we need to load the data:

Code

ss_PDC11_batch2 <- readRDS("data/ss_PDC11_batch2.rds")[-1,]
betas_PDC11_batch2 <- readRDS("data/betas_PDC11_batch2.rds")
DMPann_PDC11_batch2 <- readRDS("data/DMPann_PDC11_batch2.rds")

🛑✋ idcol variable changes for PDC11

In the idcol variable, we store the name of the probe that contains the methylation measure for a specific cg site. These probe names vary for EPIC v2 arrays but remain consistent for 450k and EPIC arrays. However, in the case of EPIC v2 arrays, we can utilize the information in the EPICv1_Loci column, which contains the equivalent cg site ID if available from the EPIC v1 array. This allows us to compare cg sites across different array types.

Code

idcol<-"EPICv1_Loci"

Then apply the function to select the top 1000 most variable sites:

Code

n = 1000                                               # change this value to select a different number of probes
betas_PDC11_batch2_heat <- top_beta(betas_PDC11_batch2,n=n)

Now we can generate the plot this using the functions defined Section 0.3:

Code

heat_list_PDC11_batch2 <- meth_heatmap(samplesheet = ss_PDC11_batch2, bvals = betas_PDC11_batch2_heat, annotation = DMPann_PDC11_batch2, idcol = idcol)

And finally save with the desired resolution and titles:

Code

library(ComplexHeatmap)
library(circlize)

ComplexHeatmap::draw(heat_list_PDC11_batch2, row_title = paste0("top ",n ," across-sample most variable sites"), row_title_gp = gpar(col = "darkblue") , height = unit(1*1/96, "inch")*n, column_title = paste0( "PDC11_batch2 cell line top ", n, " most variable Probes Heatmap"),  merge_legend = TRUE)

Combined plot:

In this case we are going to combine the 3 cell lines into the same plot so we can compare between cell lines. In order to see differences between the cell lines we could choose one of these options:

Option A: Use the top variable sites across all samples (same approach as before but using all samples now)

- Identify Top Variable Sites:
    Calculate variability for each site across all samples.
    Select the top variable sites based on this calculation.

- Generate Combined Plot:
    Plot the selected top variable sites. 
    Use different colors or symbols for each cell line.
    

Option B: Merge the top 1000 sites from each cell line

- Use the Top 1000 Sites for each Cell Line we have calculated above.
- Merge into a Combined Pool.
- Generate Combined Plot.

In this case we are going to choose option B, but I am sure you will be able to reproduce the steps to make option A 😉.

Combine the top1000 sites for each cell line into a single set of probes:

Code

l_betas_ids <- list(rownames(betas_PDC11_batch2_heat),rownames(betas_H358_heat),rownames(betas_H2009_heat))
common_cgsites <- Reduce(union,l_betas_ids)
length(common_cgsites)

[1] 2888

Trim the beta values to contain only those sites (if present) for all the cell lines:

Code

library(data.table)
l_betas <- list(betas_PDC11_batch2, betas_H358, betas_H2009)
betas_trimmed <- lapply(l_betas,function(x){
  dt <- as.data.table(x,keep.rownames = "id")
  dt <- dt[common_cgsites,,on="id"]
  return(dt)
})

Combine the beta values for all samples in a single object:

Code

common_betas <- Reduce(function(x,y)merge(x,y,by = 'id',all=T), betas_trimmed)
b<- as.matrix(common_betas[,-1])
rownames(b)<-common_betas$id
common_betas <- b[complete.cases(b),]

NA action

A total of 597 probes, out of the 2291 available, are absent for certain cell lines. This makes it impossible for the distance calculation algorithm to work since all values within the grouping factor are missing. To adress this issue this probes have been removed.

Combine annotation and samplesheet:

Code

# Add arraytype in all sample sheets:
ss_H2009[,arraytype:="EPIC"]
ss_H358$arraytype <- "EPIC"
common_ss <- rbindlist(list(ss_PDC11_batch2,ss_H2009,ss_H358),use.names=TRUE)
# Make idcol consistent for EPICv2:
DMPann_PDC11_batch2$Name <- DMPann_PDC11_batch2$EPICv1_Loci
common_annotation <- rbindlist(list(DMPann_PDC11_batch2,DMPann_H2009,DMPann_H358),use.names=TRUE, fill=TRUE)
colnames(common_betas) <- common_ss[colnames(b),Sample_Name,on="barcode"]

Modify the heatmap functions (#**):

Code

library(ComplexHeatmap)


meth_heatmap2 <- function(samplesheet, bvals, annotation, idcol="rn", sample_ids = "Sample_Name"   ){
  require(ComplexHeatmap)
  require(data.table)
  data.table::setDT(annotation)
  setkeyv(annotation,idcol)
  # Order beta values same as samplesheet:
  bvals <- bvals[,with(samplesheet,get(sample_ids))]
  
  # Annotation object for top annotation:
  data.table::setDT(samplesheet)
  # purity = samplesheet[,.SD,.SDcols=startsWith(names(samplesheet),"purity")]
  ha_column = HeatmapAnnotation(
    annotation_name_side = "left", 
    
    CL = anno_block( gp=gpar( fill=RColorBrewer::brewer.pal(n=length(unique(samplesheet$CL)) , name = "Accent")),                                                                             
                      labels = unique(samplesheet$CL), #unique(samplesheet$Type),                        #**
                      labels_gp = gpar(col = "black", fontsize = 11)),                                  #**
    Type = samplesheet$Type,

    Condition = samplesheet$Condition,
    # purity = unlist(purity),

    col = #A list of named vectors were names = vector values and value = color.
      list(
      Condition = setNames(
        palette.colors(length(unique(samplesheet$Condition)),
                       palette = "Dark"
                       ),
        unique(samplesheet$Condition)
      ),
      Type = setNames(
        rainbow(n=length(unique(samplesheet$Type))),
        unique(samplesheet$Type)
      ),
      # purity = col_fbp(x=samplesheet,bp=seq(min(purity),max(purity),length.out=5),color="green3"),
      NULL
      )
    
  )
  
  heat_list<-ComplexHeatmap::Heatmap(
    matrix = bvals, 
    #Color:
    col=col_fn(bvals),#col_fun, # Color defined in col_fun above
    na_col="grey",
    
    #Label:
    heatmap_legend_param = list(
    at = c(0, 0.5, 1),
    # labels = c("hypo", 0, "hyper"),
    title = paste0(expression(beta),"-vals"),
    legend_height = unit(4, "cm"),
    title_position = "leftcenter-rot"
    ),
    
    
    #Rows:
    show_row_names = F,
    #row_title = "Amino acids",
    row_names_side = "left",
    #left_annotation = ha_boxplot,
    # clustering_distance_rows = "manhattan",
    
    #Columns:
    show_column_names = TRUE,
    column_names_side = "top",
    column_title_side = "bottom",
    column_names_max_height = unit(4, "cm"),
    column_names_gp = gpar(fontsize = 9),
    column_names_rot = 90,
    cluster_columns=F,
    column_split = droplevels(samplesheet$CL), #factor(samplesheet$Type,levels = c("Case","Control")),       #**
    column_title = paste0(expression(beta)," values"),

    #Annotation bar:
    top_annotation = ha_column,
   
      
    
    #Aspect ratios:
    #column_dend_height=unit(4, "cm")
    heatmap_width = unit(2, "npc"),
    heatmap_height = unit(16, "cm")
  )
  
  ann<-annotation[,.(V1=unique(Relation_to_Island)),by=idcol]
  
  mat<-ann[rownames(bvals),,on=idcol][!duplicated(Name),V1]                                             #**
  heat_list<-heat_list + 
    Heatmap(
      matrix = mat, 
      name = "Relation to island",
      #Color:
      col=setNames(palette.colors(length(unique(annotation$Relation_to_Island)),palette = "Set1"),sort(unique(annotation$Relation_to_Island))),#col_fun, # Color defined in col_fun above
      #Legend:
      show_heatmap_legend = TRUE,
      show_column_names = TRUE,
      column_names_side = "top",
      column_names_max_height = unit(2, "cm"),
      column_names_gp = gpar(fontsize = 9),
      column_names_rot = 90,
      heatmap_width = unit(1, "npc"),
            
                )
} 

Generate the heatmap:

Code

heat_list_common <- meth_heatmap2(samplesheet = common_ss, bvals = common_betas, annotation = common_annotation, idcol = "Name")

`use_raster` is automatically set to TRUE for a matrix with more than
2000 rows. You can control `use_raster` argument by explicitly setting
TRUE/FALSE to it.

Set `ht_opt$message = FALSE` to turn off this message.

'magick' package is suggested to install to give better rasterization.

Set `ht_opt$message = FALSE` to turn off this message.

Adjust dimensions:

Code

library(ComplexHeatmap)
plotsizes=c(100,200)
y = NULL
for(nr in plotsizes) {
  betas<-top_beta(betas_PDC11,nr)
  
    ht = ComplexHeatmap::draw(meth_heatmap2(samplesheet = common_ss, bvals = common_betas, annotation = common_annotation, idcol = "Name"),height=unit(1/96, "inch")*nr)
    ht_height = sum(component_height(ht)) + unit(4, "mm")
    ht_height = convertHeight(ht_height, "inch", valueOnly = TRUE)
    y = c(y, ht_height)
}
sizemod <- lm(y ~ plotsizes)
sizemod

Draw and save:

Code

 png(paste0("Combined_heatmap_resized.png"),
     width =  14,
     height = sizemod$coefficients[2]*NROW(common_betas) + sizemod$coefficients[1],
     units = "in",
     res = 96
     )
ComplexHeatmap::draw(heat_list_common,
                     row_title = paste0("cg Sites"),
                     row_title_gp = gpar(col = "darkblue"),
                     height = unit(1*1/96, "inch")*NROW(common_betas),
                     column_title = paste0( "Combined Cell Lines: Top 1000 Most Variable Sites from each Cell Line"),
                     merge_legend = TRUE
                     )
dev.off()

(a) top1k_combined

Figure 3: ?(caption)